I think that if game companies can find good corpuses to train on and train their own LLMs with less strict respectability filters, using AI for NPC dialogue could be a legitimate boon to immersion. It’s one of the few cases where I could see LLMs being ethically sourced (for example using the massive amounts of text in The Elder Scrolls universe to train on) and not displacing too many jobs since there would still need to be writers for main quests and for guiding plot points on random encounters.